1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Copyright (C) 2014 Facebook. All rights reserved. 4 */ 5 6 #ifndef BTRFS_QGROUP_H 7 #define BTRFS_QGROUP_H 8 9 #include <linux/types.h> 10 #include <linux/spinlock.h> 11 #include <linux/rbtree.h> 12 #include <linux/kobject.h> 13 #include <linux/list.h> 14 #include <uapi/linux/btrfs_tree.h> 15 16 struct extent_buffer; 17 struct extent_changeset; 18 struct btrfs_delayed_extent_op; 19 struct btrfs_fs_info; 20 struct btrfs_root; 21 struct btrfs_ioctl_quota_ctl_args; 22 struct btrfs_trans_handle; 23 struct btrfs_delayed_ref_root; 24 struct btrfs_inode; 25 26 /* 27 * Btrfs qgroup overview 28 * 29 * Btrfs qgroup splits into 3 main part: 30 * 1) Reserve 31 * Reserve metadata/data space for incoming operations 32 * Affect how qgroup limit works 33 * 34 * 2) Trace 35 * Tell btrfs qgroup to trace dirty extents. 36 * 37 * Dirty extents including: 38 * - Newly allocated extents 39 * - Extents going to be deleted (in this trans) 40 * - Extents whose owner is going to be modified 41 * 42 * This is the main part affects whether qgroup numbers will stay 43 * consistent. 44 * Btrfs qgroup can trace clean extents and won't cause any problem, 45 * but it will consume extra CPU time, it should be avoided if possible. 46 * 47 * 3) Account 48 * Btrfs qgroup will updates its numbers, based on dirty extents traced 49 * in previous step. 50 * 51 * Normally at qgroup rescan and transaction commit time. 52 */ 53 54 /* 55 * Special performance optimization for balance. 56 * 57 * For balance, we need to swap subtree of subvolume and reloc trees. 58 * In theory, we need to trace all subtree blocks of both subvolume and reloc 59 * trees, since their owner has changed during such swap. 60 * 61 * However since balance has ensured that both subtrees are containing the 62 * same contents and have the same tree structures, such swap won't cause 63 * qgroup number change. 64 * 65 * But there is a race window between subtree swap and transaction commit, 66 * during that window, if we increase/decrease tree level or merge/split tree 67 * blocks, we still need to trace the original subtrees. 68 * 69 * So for balance, we use a delayed subtree tracing, whose workflow is: 70 * 71 * 1) Record the subtree root block get swapped. 72 * 73 * During subtree swap: 74 * O = Old tree blocks 75 * N = New tree blocks 76 * reloc tree subvolume tree X 77 * Root Root 78 * / \ / \ 79 * NA OB OA OB 80 * / | | \ / | | \ 81 * NC ND OE OF OC OD OE OF 82 * 83 * In this case, NA and OA are going to be swapped, record (NA, OA) into 84 * subvolume tree X. 85 * 86 * 2) After subtree swap. 87 * reloc tree subvolume tree X 88 * Root Root 89 * / \ / \ 90 * OA OB NA OB 91 * / | | \ / | | \ 92 * OC OD OE OF NC ND OE OF 93 * 94 * 3a) COW happens for OB 95 * If we are going to COW tree block OB, we check OB's bytenr against 96 * tree X's swapped_blocks structure. 97 * If it doesn't fit any, nothing will happen. 98 * 99 * 3b) COW happens for NA 100 * Check NA's bytenr against tree X's swapped_blocks, and get a hit. 101 * Then we do subtree scan on both subtrees OA and NA. 102 * Resulting 6 tree blocks to be scanned (OA, OC, OD, NA, NC, ND). 103 * 104 * Then no matter what we do to subvolume tree X, qgroup numbers will 105 * still be correct. 106 * Then NA's record gets removed from X's swapped_blocks. 107 * 108 * 4) Transaction commit 109 * Any record in X's swapped_blocks gets removed, since there is no 110 * modification to the swapped subtrees, no need to trigger heavy qgroup 111 * subtree rescan for them. 112 */ 113 114 /* 115 * These flags share the flags field of the btrfs_qgroup_status_item with the 116 * persisted flags defined in btrfs_tree.h. 117 * 118 * To minimize the chance of collision with new persisted status flags, these 119 * count backwards from the MSB. 120 */ 121 #define BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN (1ULL << 63) 122 #define BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING (1ULL << 62) 123 124 #define BTRFS_QGROUP_DROP_SUBTREE_THRES_DEFAULT (3) 125 126 /* 127 * Record a dirty extent, and info qgroup to update quota on it 128 */ 129 struct btrfs_qgroup_extent_record { 130 /* 131 * The bytenr of the extent is given by its index in the dirty_extents 132 * xarray of struct btrfs_delayed_ref_root left shifted by 133 * fs_info->sectorsize_bits. 134 */ 135 136 u64 num_bytes; 137 138 /* 139 * For qgroup reserved data space freeing. 140 * 141 * @data_rsv_refroot and @data_rsv will be recorded after 142 * BTRFS_ADD_DELAYED_EXTENT is called. 143 * And will be used to free reserved qgroup space at 144 * transaction commit time. 145 */ 146 u32 data_rsv; /* reserved data space needs to be freed */ 147 u64 data_rsv_refroot; /* which root the reserved data belongs to */ 148 struct ulist *old_roots; 149 }; 150 151 struct btrfs_qgroup_swapped_block { 152 struct rb_node node; 153 154 int level; 155 bool trace_leaf; 156 157 /* bytenr/generation of the tree block in subvolume tree after swap */ 158 u64 subvol_bytenr; 159 u64 subvol_generation; 160 161 /* bytenr/generation of the tree block in reloc tree after swap */ 162 u64 reloc_bytenr; 163 u64 reloc_generation; 164 165 u64 last_snapshot; 166 struct btrfs_key first_key; 167 }; 168 169 /* 170 * Qgroup reservation types: 171 * 172 * DATA: 173 * space reserved for data 174 * 175 * META_PERTRANS: 176 * Space reserved for metadata (per-transaction) 177 * Due to the fact that qgroup data is only updated at transaction commit 178 * time, reserved space for metadata must be kept until transaction 179 * commits. 180 * Any metadata reserved that are used in btrfs_start_transaction() should 181 * be of this type. 182 * 183 * META_PREALLOC: 184 * There are cases where metadata space is reserved before starting 185 * transaction, and then btrfs_join_transaction() to get a trans handle. 186 * Any metadata reserved for such usage should be of this type. 187 * And after join_transaction() part (or all) of such reservation should 188 * be converted into META_PERTRANS. 189 */ 190 enum btrfs_qgroup_rsv_type { 191 BTRFS_QGROUP_RSV_DATA, 192 BTRFS_QGROUP_RSV_META_PERTRANS, 193 BTRFS_QGROUP_RSV_META_PREALLOC, 194 BTRFS_QGROUP_RSV_LAST, 195 }; 196 197 /* 198 * Represents how many bytes we have reserved for this qgroup. 199 * 200 * Each type should have different reservation behavior. 201 * E.g, data follows its io_tree flag modification, while 202 * *currently* meta is just reserve-and-clear during transaction. 203 * 204 * TODO: Add new type for reservation which can survive transaction commit. 205 * Current metadata reservation behavior is not suitable for such case. 206 */ 207 struct btrfs_qgroup_rsv { 208 u64 values[BTRFS_QGROUP_RSV_LAST]; 209 }; 210 211 /* 212 * one struct for each qgroup, organized in fs_info->qgroup_tree. 213 */ 214 struct btrfs_qgroup { 215 u64 qgroupid; 216 217 /* 218 * state 219 */ 220 u64 rfer; /* referenced */ 221 u64 rfer_cmpr; /* referenced compressed */ 222 u64 excl; /* exclusive */ 223 u64 excl_cmpr; /* exclusive compressed */ 224 225 /* 226 * limits 227 */ 228 u64 lim_flags; /* which limits are set */ 229 u64 max_rfer; 230 u64 max_excl; 231 u64 rsv_rfer; 232 u64 rsv_excl; 233 234 /* 235 * reservation tracking 236 */ 237 struct btrfs_qgroup_rsv rsv; 238 239 /* 240 * lists 241 */ 242 struct list_head groups; /* groups this group is member of */ 243 struct list_head members; /* groups that are members of this group */ 244 struct list_head dirty; /* dirty groups */ 245 246 /* 247 * For qgroup iteration usage. 248 * 249 * The iteration list should always be empty until qgroup_iterator_add() 250 * is called. And should be reset to empty after the iteration is 251 * finished. 252 */ 253 struct list_head iterator; 254 255 /* 256 * For nested iterator usage. 257 * 258 * Here we support at most one level of nested iterator calls like: 259 * 260 * LIST_HEAD(all_qgroups); 261 * { 262 * LIST_HEAD(local_qgroups); 263 * qgroup_iterator_add(local_qgroups, qg); 264 * qgroup_iterator_nested_add(all_qgroups, qg); 265 * do_some_work(local_qgroups); 266 * qgroup_iterator_clean(local_qgroups); 267 * } 268 * do_some_work(all_qgroups); 269 * qgroup_iterator_nested_clean(all_qgroups); 270 */ 271 struct list_head nested_iterator; 272 struct rb_node node; /* tree of qgroups */ 273 274 /* 275 * temp variables for accounting operations 276 * Refer to qgroup_shared_accounting() for details. 277 */ 278 u64 old_refcnt; 279 u64 new_refcnt; 280 281 /* 282 * Sysfs kobjectid 283 */ 284 struct kobject kobj; 285 }; 286 287 /* Glue structure to represent the relations between qgroups. */ 288 struct btrfs_qgroup_list { 289 struct list_head next_group; 290 struct list_head next_member; 291 struct btrfs_qgroup *group; 292 struct btrfs_qgroup *member; 293 }; 294 295 struct btrfs_squota_delta { 296 /* The fstree root this delta counts against. */ 297 u64 root; 298 /* The number of bytes in the extent being counted. */ 299 u64 num_bytes; 300 /* The generation the extent was created in. */ 301 u64 generation; 302 /* Whether we are using or freeing the extent. */ 303 bool is_inc; 304 /* Whether the extent is data or metadata. */ 305 bool is_data; 306 }; 307 308 static inline u64 btrfs_qgroup_subvolid(u64 qgroupid) 309 { 310 return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1)); 311 } 312 313 /* 314 * For qgroup event trace points only 315 */ 316 enum { 317 ENUM_BIT(QGROUP_RESERVE), 318 ENUM_BIT(QGROUP_RELEASE), 319 ENUM_BIT(QGROUP_FREE), 320 }; 321 322 enum btrfs_qgroup_mode { 323 BTRFS_QGROUP_MODE_DISABLED, 324 BTRFS_QGROUP_MODE_FULL, 325 BTRFS_QGROUP_MODE_SIMPLE 326 }; 327 328 enum btrfs_qgroup_mode btrfs_qgroup_mode(const struct btrfs_fs_info *fs_info); 329 bool btrfs_qgroup_enabled(const struct btrfs_fs_info *fs_info); 330 bool btrfs_qgroup_full_accounting(const struct btrfs_fs_info *fs_info); 331 int btrfs_quota_enable(struct btrfs_fs_info *fs_info, 332 struct btrfs_ioctl_quota_ctl_args *quota_ctl_args); 333 int btrfs_quota_disable(struct btrfs_fs_info *fs_info); 334 int btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info); 335 void btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info); 336 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info, 337 bool interruptible); 338 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, u64 dst, 339 struct btrfs_qgroup_list *prealloc); 340 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, 341 u64 dst); 342 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid); 343 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid); 344 int btrfs_qgroup_cleanup_dropped_subvolume(struct btrfs_fs_info *fs_info, u64 subvolid); 345 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid, 346 struct btrfs_qgroup_limit *limit); 347 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info); 348 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info); 349 350 int btrfs_qgroup_trace_extent_nolock( 351 struct btrfs_fs_info *fs_info, 352 struct btrfs_delayed_ref_root *delayed_refs, 353 struct btrfs_qgroup_extent_record *record, 354 u64 bytenr); 355 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans, 356 struct btrfs_qgroup_extent_record *qrecord, 357 u64 bytenr); 358 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr, 359 u64 num_bytes); 360 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans, 361 struct extent_buffer *eb); 362 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans, 363 struct extent_buffer *root_eb, 364 u64 root_gen, int root_level); 365 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr, 366 u64 num_bytes, struct ulist *old_roots, 367 struct ulist *new_roots); 368 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans); 369 int btrfs_run_qgroups(struct btrfs_trans_handle *trans); 370 int btrfs_qgroup_check_inherit(struct btrfs_fs_info *fs_info, 371 struct btrfs_qgroup_inherit *inherit, 372 size_t size); 373 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid, 374 u64 objectid, u64 inode_rootid, 375 struct btrfs_qgroup_inherit *inherit); 376 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info, 377 u64 ref_root, u64 num_bytes, 378 enum btrfs_qgroup_rsv_type type); 379 380 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 381 int btrfs_verify_qgroup_counts(const struct btrfs_fs_info *fs_info, u64 qgroupid, 382 u64 rfer, u64 excl); 383 #endif 384 385 /* New io_tree based accurate qgroup reserve API */ 386 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode, 387 struct extent_changeset **reserved, u64 start, u64 len); 388 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len, u64 *released); 389 int btrfs_qgroup_free_data(struct btrfs_inode *inode, 390 struct extent_changeset *reserved, u64 start, 391 u64 len, u64 *freed); 392 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes, 393 enum btrfs_qgroup_rsv_type type, bool enforce); 394 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes, 395 enum btrfs_qgroup_rsv_type type, bool enforce, 396 bool noflush); 397 /* Reserve metadata space for pertrans and prealloc type */ 398 static inline int btrfs_qgroup_reserve_meta_pertrans(struct btrfs_root *root, 399 int num_bytes, bool enforce) 400 { 401 return __btrfs_qgroup_reserve_meta(root, num_bytes, 402 BTRFS_QGROUP_RSV_META_PERTRANS, 403 enforce, false); 404 } 405 static inline int btrfs_qgroup_reserve_meta_prealloc(struct btrfs_root *root, 406 int num_bytes, bool enforce, 407 bool noflush) 408 { 409 return __btrfs_qgroup_reserve_meta(root, num_bytes, 410 BTRFS_QGROUP_RSV_META_PREALLOC, 411 enforce, noflush); 412 } 413 414 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes, 415 enum btrfs_qgroup_rsv_type type); 416 417 /* Free per-transaction meta reservation for error handling */ 418 static inline void btrfs_qgroup_free_meta_pertrans(struct btrfs_root *root, 419 int num_bytes) 420 { 421 __btrfs_qgroup_free_meta(root, num_bytes, 422 BTRFS_QGROUP_RSV_META_PERTRANS); 423 } 424 425 /* Pre-allocated meta reservation can be freed at need */ 426 static inline void btrfs_qgroup_free_meta_prealloc(struct btrfs_root *root, 427 int num_bytes) 428 { 429 __btrfs_qgroup_free_meta(root, num_bytes, 430 BTRFS_QGROUP_RSV_META_PREALLOC); 431 } 432 433 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root); 434 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes); 435 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode); 436 437 /* btrfs_qgroup_swapped_blocks related functions */ 438 void btrfs_qgroup_init_swapped_blocks( 439 struct btrfs_qgroup_swapped_blocks *swapped_blocks); 440 441 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root); 442 int btrfs_qgroup_add_swapped_blocks(struct btrfs_root *subvol_root, 443 struct btrfs_block_group *bg, 444 struct extent_buffer *subvol_parent, int subvol_slot, 445 struct extent_buffer *reloc_parent, int reloc_slot, 446 u64 last_snapshot); 447 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans, 448 struct btrfs_root *root, struct extent_buffer *eb); 449 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans); 450 bool btrfs_check_quota_leak(const struct btrfs_fs_info *fs_info); 451 int btrfs_record_squota_delta(struct btrfs_fs_info *fs_info, 452 const struct btrfs_squota_delta *delta); 453 454 #endif 455